Delete DataManager #691
Delete DataManager #691
Conversation
Also done: - Create a forwardInternal with labels and the implementations in the higher level blocks SequentialBlock and ParallelBlock - Removed duplicate testSeqToSeq test Change-Id: I16f5343886a62538f3a247df062d28a765805cb8
|
LGTM when I see the title. Look into the code now |
| .addSingleton( | ||
| a -> { | ||
| try { | ||
| return embedding.embedText(a); |
There was a problem hiding this comment.
The problem to put it here is: I would like to freeze this block and not use for training. If remove DataManager, I will not be able to freeze the block
There was a problem hiding this comment.
This would be frozen because the ModelZooTextEmbedding freezes (it uses a separate predictor). In a different case, you should be able to just call embedding.freeze() to freeze just the embedding.
There was a problem hiding this comment.
Can you verify if PyTorch engine will work in this case? I am not sure if the training work are correctly handled
There was a problem hiding this comment.
There is no word embedding for pt in the model zoo so I can't test it. But, is there a reason that PT may not work?
There was a problem hiding this comment.
It should work by default. Just concerned if the NDArray inside will be impacted. In the DistiledBert training for Amazon Dataset, we throw the embedding layer into a lambda function to avoid parameter collections
Also done:
blocks SequentialBlock and ParallelBlock